Quantitative Imaging and Tomography with Polychromatic X-rays

Abstract

X-ray computed tomography (XCT) is a powerful tool, renowned for its ability to non-destructively produce highly-detailed 3D images of an object's internal structure. The scanning process captures X-ray projections of the object of interest from various angles and then computationally creates a volumetric map of its X-ray attenuation coefficient. Conventional lab-based XCT invariably employs polychromatic (or broadband) X-ray radiation. This causes imaging artifacts since attenuation (the reconstructed quantity) is a function of X-ray energy, making it difficult to differentiate materials, let alone obtain quantitative information. While some methods have been proposed in the literature, no effective approach has been widely applied. In this work, we propose a two-material (2M) model for polychromatic multi-energy XCT. This model is an alternative to the well-known Alvarez-Macovski (AM) model, that approximates that X-ray attenuation for a material by a simple function of its density, atomic number, and the energy of X-rays. The 2M model assumes that X-ray attenuation for a material can be modeled by a linear combination of that of two known materials (basis materials). We can then effectively reconstruct a quantitative estimate of the material properties: density and atomic number, from those of the two basis materials. Moreover, since these models capture the energy-dependence of attenuation, they inherently account for common tomography artifacts. In simulation tomography, the 2M model showed improved artifact correction and yielded quantitative results comparable to the AM model, with densities and atomic numbers having a relative error within 10\%. Through experiments, we demonstrated that the 2M model can simplify laboratory XCT calibration. Unlike the AM model, which requires full spectrum knowledge, the 2M approach only needs measurements of X-ray attenuation for specific basis material combinations to approximate where our target material fits. While the 2M model exhibited some potential, we further identified its limitations in the lab and suggested future improvements for its mature implementation Show more